next up previous contents
Next: The Hausdorff Distance Up: Shape Measurement and Defect Previous: Shape Defect Detection Problem

   
Matching Method for Defect Detection

Our approach is based on the notion of the Hausdorff distance (HD) between two point sets, which is introduced in section 3.1. This distance proved to be a general, efficient and robust measure of similarity between two shapes. Its robustness to noisy and incomplete data stems from the global distance between two point sets it operates with: no correspondences between particular points are searched for. A survey of tasks and methods related to the shape matching using the Hausdorff distance can be found in [5].

We apply the HD to a new problem, that of shape defect detection, and present an algorithm for fast computation of the reference pose based on the distance transform [1]. The distance transform (DT) is discussed in section 3.2.

The proposed method first gives an initial guess that approximately normalizes the pose of the measured shape, then uses the modified median Hausdorff distance as the precise measure of shape correspondence. The main steps of the method are described in section 3.3.

Several options had been considered to obtain the initial matching. Any local feature based criterion, e.g., the coincidence of the baselines of the shapes, would be neither universal nor stable, because defects may distort that particular local feature. As the centroid is less sensitive to local defects, we finally decided to initially normalize the position by superimposing the centroids of the two shapes. This solution is often used in the industrial machine vision practice.



 
next up previous contents
Next: The Hausdorff Distance Up: Shape Measurement and Defect Previous: Shape Defect Detection Problem
Dmitry Chetverikov
1998-11-16